Overview
Peter Stainberger, creator of PSPDFkit (used on 1 billion+ devices) and now Clawbot, shares how AI has fundamentally changed his software development workflow. He explains why he no longer reads most code he ships yet maintains quality through architectural thinking and validation loops. After burning out from his PDF company, taking a 3-year break, and returning to find AI transformed development, he now merges 600 commits daily using AI agents while focusing on system design rather than code implementation.
Key Takeaways
- Focus on architectural thinking over code implementation - AI handles the plumbing while you design the system structure and user experience
- Close the validation loop - AI excels at coding because code can be compiled, tested, and verified unlike creative writing; design your systems to enable automated testing
- Embrace parallel agent management - Modern development means orchestrating multiple AI agents simultaneously, requiring mental context switching skills similar to chess grandmasters playing multiple boards
- Shift from code reviews to prompt reviews - Understanding how someone solved a problem (their prompts) provides higher signal than reviewing the generated code itself
- Design for agent consumption - Structure codebases and workflows to be easily navigated by AI agents, not just human developers
Topics Covered
- 0:00 - Introduction and Clawbot explosion: Peter’s background creating PSPDFkit and current project Clawbot gaining rapid community traction
- 2:00 - Early tech journey and entrepreneurship: Growing up in rural Austria, first programming experiences, and building a dating app that made $10k in first month
- 10:00 - Building PSPDFkit - the PDF problem: How a freelance project to fix a magazine app led to building a PDF framework used on billions of devices
- 18:00 - Scaling PSPDFkit company culture: Building remote-first team of 70+ people, technical blogging strategy, and enterprise sales challenges
- 30:00 - Burnout and 3-year break from tech: CEO stress, technical debt, team conflicts leading to selling shares and complete tech detox
- 35:00 - AI awakening with Claude and agents: Returning to find AI transformed development, first experiences with Claude Code and getting ‘AI pilled’
- 43:00 - New development workflow with AI: Managing multiple AI agents in parallel, architectural thinking vs code reading, and ‘weaving’ code instead of writing
- 57:00 - Closing the loop principle: Why AI excels at code but struggles with writing - the importance of validation loops and automated testing
- 1:13:00 - Building Clawbot personal assistant: Creating a hyper-personal AI assistant that can control computers, make calls, and integrate with daily life
- 1:35:00 - Future of software engineering teams: How companies will need to restructure, the death of code reviews, and prompt requests replacing pull requests
- 1:45:00 - Advice for new developers: Recommendations for entering the field in an AI-dominated world and building curiosity-driven learning